Skip to content

copilot-guide.md

Latest commit

 

History

History
77 lines (57 loc) · 3.37 KB

copilot-guide.md

File metadata and controls

77 lines (57 loc) · 3.37 KB
Raw

Copilot Guide

Copilot support is available through two layers:

  • prompt handoff files written into .wi/runtime/
  • custom phase agents installed from .agents/ into your local .copilot/agents/ directory

Install the AWF phase agents

PowerShell:

./installers/install-copilot-agents.ps1

Bash:

./installers/install-copilot-agents.sh

By default the installers copy *.agent.md files from .agents/ into ~/.copilot/agents.

Included agents

  • awf-intake
  • awf-planner
  • awf-implementer
  • awf-reviewer
  • awf-verifier

These names line up with .wi/config.json -> phase_execution.<phase>.agent.

Recommended phase mapping

{
  "phase_execution": {
    "intake": { "adapter": "copilot", "model": "gpt-5", "agent": "awf-intake" },
    "planning": { "adapter": "copilot", "model": "gpt-5", "agent": "awf-planner" },
    "implementation": { "adapter": "claude", "model": "claude-sonnet-4-5", "agent": "awf-implementer" },
    "review": { "adapter": "gemini", "model": "gemini-2.5-pro", "agent": "awf-reviewer" },
    "final_verification": { "adapter": "copilot", "model": "gpt-5", "agent": "awf-verifier" }
  }
}

Optional auto-launch

If you have a scriptable GitHub Copilot CLI setup, configure the launch command that matches your machine.

For environments that expose Copilot through the standalone copilot CLI:

{
  "adapters": {
    "copilot": {
      "launch_command": ["copilot", "--interactive", "{{PROMPT_TEXT}}", "--autopilot", "--allow-all", "--add-dir", "{{WORKING_DIRECTORY}}", "--model", "{{MODEL}}"]
    }
  }
}

This shape uses Copilot's interactive prompt launch, grants the working directory up front, and opts into autonomous permission handling. AWF omits --model automatically until a model has been chosen for the phase.

awf tooling-status and awf doctor now treat Copilot as installed when copilot is available on PATH.

How the routing works

  • AWF writes the active implementation handoff file to .wi/runtime/copilot-handoff.md when implementation.adapter is copilot.
  • The runtime packet also includes the selected adapter, model, and phase agent so the execution choice is visible inside the repo.
  • For planning, review, and verification, the phase routing config serves as the contract for which Copilot custom agent you should pick when opening that phase manually.
  • For intake, awf start-story --interactive or awf clarify-story --interactive lets you resolve blocking clarification questions in one terminal session before handing off to the configured intake or planning agent.
  • Clarification questions are generated from the ticket content itself, and the interactive loop lets you review and revise answers before continuing.
  • awf start-story --ai now generates an intake handoff for the configured phase_execution.intake adapter, so Copilot can refine .wi/story.json before AWF validates it.
  • Use --design-file <path> or --design-note <text> when intake should compare the story against design context before it asks questions.
  • AI-assisted intake now persists .wi/intake-context.json, and awf clarify-story --ai can reuse that same story plus design grounding for follow-up clarification passes.
  • In the shell, /configure can set Copilot by stage, /flow shows the current workflow step, and /stage lets you move the workflow back to planning, review, or verification when needed.